[clang] [clang] Fix couroutine error for operator new. (PR #215619)

[Oliver Hunt via cfe-commits]

ojhunt wrote:

In this particular case the switch to revert seemed unreasonably fast -- my general approach is to bring in the work around or reversion downstream to unblock our downstream bots while the appropriate solution is discussed here. 

There have been plenty of times where changes here have broken our downstream forks, and the cost of a downstream work around while a thorough discussion of the actual issue and appropriate fix in upstream has resulted in less work for everyone involved and less revert dancing.

My _personal_ view is that it is unreasonable for a downstream fork to force an immediate reversion of a change that does not break the main repository, and that's what I'm going to stick with.

But to be clear: I think that "downstream breakage that is unambiguously an error should trigger a revert" (as was the case here) is an equally reasonable position for someone to take.

https://github.com/llvm/llvm-project/pull/215619